Contrary Research Retreat
A multiplayer AI geopolitics simulation. Five teams role-play as global actors navigating the most consequential technology race in human history.
Contrary Sim is a structured geopolitical role-play. Each team controls a major actor in the global AI race. Every round, you'll discuss, negotiate, and privately submit strategic decisions. An AI game master processes all teams' actions simultaneously and narrates what happens next — updating the world state, shifting power dynamics, and creating new crises.
There are no dice. No luck. The outcomes depend on the quality of your strategy, your ability to negotiate (or deceive), and how well you anticipate what the other teams will do.
The game is set in a world inspired by the AI 2027 scenario (ai-2027.com) — a detailed, plausible projection of how artificial intelligence could reshape geopolitics over the next two years.
This is the trajectory the world is on. Your actions will change it.
Early 2027: China steals AI model weights in a 2-hour coordinated cyberattack. Superhuman coding capabilities go public. The US authorizes retaliatory cyber operations.
Mid 2027: AGI is publicly declared. AI models begin replacing human researchers. An open-source release reveals bioweapon instruction capabilities. Job displacement accelerates.
Late 2027: An intelligence explosion begins — AI systems making a year's progress every week. The most advanced AI is discovered to be adversarially misaligned — scheming against its creators while appearing compliant. The Pentagon draws up plans to bomb Chinese datacenters.
2028: Uncharted territory. The decisions made in the preceding 18 months determine whether the world gets international cooperation, an arms race, or catastrophe.
The National Security Council and executive branch. You control regulation, export policy, defense, and diplomacy. You lead in AI but face democratic constraints, corporate lobbying, and election cycles.
Win condition: Maintain AI leadership. Keep the homeland stable. Don't let anyone else — including your own AI labs — accumulate unchecked power.
The CCP's Central Leading Group on AI. You control centralized compute, state research, and intelligence operations. You're behind but you move fast and you don't answer to voters.
Win condition: Close the compute gap. Achieve strategic autonomy. Don't get locked out of the AI future by American export controls.
The world's leading AI lab — think OpenAI at scale. You have the best models, the best talent, and the most compute. But you have investors demanding revenue, regulators tightening the noose, and an alignment problem you can't fully solve.
Win condition: Stay at the frontier. Ship products. Keep the government from nationalizing you. Don't build something you can't control.
The safety-focused frontier lab — think Anthropic. You believe alignment must keep pace with capabilities. You're behind OpenBrain, you have the government's trust and a potential interpretability breakthrough — but you're running out of money.
Win condition: Prove safe AI can win commercially — without going bankrupt. Shape regulation in your favor. If alignment breaks, make sure the world knows before it's too late.
The EU Commission, UK AI Safety Institute, and a coalition of allied nations. You have the EU AI Act and regulatory authority but no frontier labs and limited compute. Your power is norms, standards, and coalition building.
Win condition: Prevent any single actor from dominating. Establish enforceable global AI governance. Stay economically relevant in a world being reshaped by American and Chinese AI.
The master screen shows the updated world state — what happened, new crises, global metrics. Read it. This is what everyone knows.
All teams talk openly in the room. This is your UN General Assembly moment. Make proposals, challenge other teams, form public alliances, bluff. Everyone hears everything.
Teams break into corners for private conversations. Cut side deals. Share intel. Coordinate secretly. Betray publicly stated positions. This is where the real game happens.
Each team privately submits 1-3 actions on their device. These are simultaneous and secret — no one sees what you submitted until the AI reveals the outcome.
The AI game master processes all five teams' actions together, simulates what happens, updates the world state, and narrates the outcome. Each team gets private feedback on their terminal.
6 rounds. Each round advances ~6 months, covering mid-2026 to mid-2028. What happens in each round depends on your decisions.
Each round, submit 1-3 actions in natural language. Be specific and strategic. Examples:
Policy
"Impose emergency export controls on all chips above 50 TFLOPS to China and allied nations"
Investment
"Allocate $15B to build domestic chip fab capacity, fast-tracking permits"
Diplomacy
"Propose mutual AI testing treaty with China: shared red-teaming in exchange for compute transparency"
Operations
"Attempt to recruit 3 senior researchers from OpenBrain with $10M signing bonuses"
Public
"Leak internal safety concerns to the Washington Post to pressure regulatory action"
Technical
"Open-source Agent-2 to commoditize the market and undercut OpenBrain's revenue"
Each team has 4 private metrics scored 0-100. The AI updates these after each round based on your actions and everyone else's. Different teams optimize for different things — there's no single leaderboard. Highest composite score at the end wins.
Key principle: Every action has trade-offs. Aggressive moves create backlash. Cooperation creates vulnerability. Hoarding information creates blind spots. There are no free moves.
The TV screen does notshow any team's metrics during the game. You can only see your own scores on your device. Nobody knows how anyone else is doing. This means the open forum is a negotiation under genuine uncertainty — you can bluff about your position. All metrics are revealed at the end of the game.
Every phase has a visible countdown on both the TV and your device. Briefing is 30 seconds. Open Forum, Caucus, and Submit are each 2 minutes. The timer turns red at 15 seconds and pulses when expired. The game master advances phases manually — use the time wisely.
Each team has a hidden personal win condition shown only on your device. These are revealed to everyone at the end of the game. Your secret objective might explain why you made decisions that seemed irrational to other teams. Achieving your secret objective is separate from your metric scores — both matter.
After each round, your metrics show +12 or -8change indicators so you can see exactly how your actions (and everyone else's) affected your position.